【原创】Erlang 之 erl_crash.dump 文件分析

      先看一下坚强兄的博文 《 [Erlang 0057] Erlang 排错利器: Erlang Crash Dump Viewer 》 中说了什么： 
1. 基于 crashdump_viewer 的 web 页面进行查看 

crashdump_viewer:start().

补充： 

The Crashdump Viewer is an HTML based tool for browsing Erlang crashdumps. Crashdump Viewer runs under the WebTool application.

2. 基于 recon 的 erl_crashdump_analyzer.sh 分析脚本 进行查看 

-=-=-=-=- 我是88界奥斯卡颁奖礼的分隔线 -=-=-=-=- 

文章中设计到两种工具，下面分别说明 

【crashdump_viewer】 

【 erl_crashdump_analyzer.sh 】 

      该工具是 recon 工具中的其中一个，详细信息可以 参考其 github 上的说明。 

      下面给出 erl_crashdump_analyzer.sh 内容的注释说明 。 

[root@YOYO Erlang]# vi erl_crashdump_analyzer.sh 

#!/usr/bin/env bash
DUMP=$1

echo -e "analyzing $DUMP, generated on: " `head -2 $DUMP | tail -1` "\n"   -- echo -e 表示接受 \ 转义；

### SLOGAN ###
grep Slogan: $DUMP -m 1         -- 仅过滤出一条（即第一条）含有 "Slogan:" 的信息

### MEMORY ###                  -- 内存使用情况统计
echo -e "\nMemory:\n==="
M=`grep -m 1 'processes' $DUMP | sed "s/processes: //"`    -- 获取 "processes: " 后的数值（字节）
let "m=$M/(1024*1024)"
echo "  processes: $m Mb"
M=`grep -m 1 'processes_used' $DUMP | sed "s/processes_used: //"`
let "m=$M/(1024*1024)"
echo "  processes_used: $m Mb"
M=`grep -m 1 'system' $DUMP | sed "s/system: //"`
let "m=$M/(1024*1024)"
echo "  system: $m Mb"
M=`grep -m 1 'atom' $DUMP | sed "s/atom: //"`
let "m=$M/(1024*1024)"
echo "  atom: $m Mb"
M=`grep -m 1 'atom_used' $DUMP | sed "s/atom_used: //"`
let "m=$M/(1024*1024)"
echo "  atom_used: $m Mb"
M=`grep -m 1 'binary' $DUMP | sed "s/binary: //"`
let "m=$M/(1024*1024)"
echo "  binary: $m Mb"
M=`grep -m 1 'code' $DUMP | sed "s/code: //"`
let "m=$M/(1024*1024)"
echo "  code: $m Mb"
M=`grep -m 1 'ets' $DUMP | sed "s/ets: //"`
let "m=$M/(1024*1024)"
echo "  ets: $m Mb"
M=`grep -m 1 'total' $DUMP | sed "s/total: //"`
let "m=$M/(1024*1024)"
echo -e "  ---\n  total: $m Mb"

### PROCESS MESSAGE QUEUES LENGTHS ###
echo -e "\nDifferent message queue lengths (5 largest different):\n==="
grep 'Message queue len' $DUMP | sed 's/Message queue length: //g' | sort -n -r | uniq -c | head -5   -- 排序所有进程的消息队列长度，取消息数最多的 5 个显示出来

### ERROR LOGGER QUEUE LENGTH ###
echo -e "\nError logger queue length:\n==="
grep -C 10 'Name: error_logger' $DUMP -m 1| grep 'Message queue length' | sed 's/Message queue length: //g'  -- 匹配第一条 "Name: error_logger" 行，同时输出其上下各 10 行数据，从这些数据中匹配 'Message queue length: ' 获取其对应值

### PORT/FILE DESCRIPTOR INFO ###
echo -e "\nFile descriptors open:\n==="
echo -e "  UDP: "   `grep 'Port controls linked-in driver:' $DUMP | grep 'udp_inet' | wc -l`  -- udp_inet 对应 UDP 端口使用
echo -e "  TCP: "   `grep 'Port controls linked-in driver:' $DUMP | grep 'tcp_inet' | wc -l`  -- tcp_inet 对应 UDP 端口使用
echo -e "  Files: " `grep 'Port controls linked-in driver:' $DUMP | grep -vi 'udp_inet' | grep -vi 'tcp_inet' | wc -l`  -- 去除 udp_inet 和 tcp_inet 对应的行，剩下的对应 FILE 相关使用（efile 或 tty_sl -c -e 等）
echo -e "  ---\n  Total: " `grep 'Port controls linked-in driver:' $DUMP | wc -l`

### NUMBER OF PROCESSES ###
echo -e "\nNumber of processes:\n==="
grep '=proc:' $DUMP | wc -l     -- 统计进程数量（以 "=proc:" 开头的部分对应进程信息开始）

### PROC HEAPS+STACK ###
echo -e "\nProcesses Heap+Stack memory sizes (words) used in the VM (5 largest different):\n==="
grep 'Stack+heap' $DUMP | sed "s/Stack+heap: //g" | sort -n -r | uniq -c | head -5              -- 获取 "Stack+heap: " 字段后的数值，排序后取前 5 

### PROC OLDHEAP ###
echo -e "\nProcesses OldHeap memory sizes (words) used in the VM (5 largest different):\n==="
grep 'OldHeap' $DUMP | sed "s/OldHeap: //g" | sort -n -r | uniq -c | head -5

### PROC STATES ###
echo -e "\nProcess States when crashing (sum): \n==="
grep 'State: ' $DUMP | sed "s/State: //g" | sort | uniq -c               -- 获取发生 crash 时的全部进程状态（例如 Waiting 或 Running）

分析举例 

[root@YOYO Erlang]# ./erl_crashdump_analyzer.sh erl_crash.dump
analyzing erl_crash.dump, generated on:  Thu Oct 8 09:51:53 2015 

Slogan: init terminating in do_boot ()

Memory:
===
  processes: 13 Mb
  processes_used: 13 Mb
  system: 24 Mb
  atom: 0 Mb
  atom_used: 0 Mb
  binary: 0 Mb
  code: 18 Mb
  ets: 0 Mb
  ---
  total: 38 Mb

Different message queue lengths (5 largest different):
===
     43 0

Error logger queue length:
===
0

File descriptors open:
===
  UDP:  0
  TCP:  2
  Files:  4
  ---
  Total:  6

Number of processes:
===
43

Processes Heap+Stack memory sizes (words) used in the VM (5 largest different):
===
      4 6772
      3 4185
      2 2586
      3 1598
      5 987

Processes OldHeap memory sizes (words) used in the VM (5 largest different):
===
      1 46422
      1 17731
      2 10958
      3 4185
      1 2586

Process States when crashing (sum):
===
      1 Running
     42 Waiting
[root@YOYO Erlang]#

从上面的输出中，可以看到

• 内存的分配量和使用量
• 消息队列信息
• 文件描述符使用信息
• 堆+栈内存使用量

补充：
      在 lib/observer-2.0/priv/bin 路径下还有一个 cdv 脚本封装了针对 crashdump_viewer 的调用